Segmentation evaluation after border thinning not working

I am currently working on a image segmentation project with data from the ISBI-2012 segmentation challenge. I am trying to run the BeanShell script at this link:

on my own predicted segmentation images. However, I cannot seem to reproduce the results with my images.

0_predict 0_base

(LEFT is predicted, RIGHT is ground truth)

I get this as a result:

I understand that the generated cell labels must be in a certain format, as described on the ISBI-2012 site:

The results are expected to be submitted as a 32-bit TIFF 3D image, which values between 0 (100% membrane certainty) and 1 (100% non-membrane certainty).

However, I have trouble understanding how exactly to save my images in this format. Is there any way to fix this issue?

Hello @ericj0114

You need indeed to convert your image to 32-bit and make the values lie in the range [0.0 1.0].

You might do that in the script by simply adding two lines after line 59:

IJ.run( proposedLabels, "32-bit", "" );
IJ.run( proposedLabels, "Enhance Contrast...", "saturated=0.3 normalize process_all");

The ground truth image should be binary though.

1 Like